Symbolic document image compression based on pattern matching techniques

Abstract

In this paper, a novel compression algorithm for Chinese document images is proposed. Initially, documents are segmented into readable components such as characters and punctuation marks. Similar patterns within the text are found by shape context matching and grouped to form a set of prototype symbols. Text redundancies can be removed by replacing repeated symbols by their corresponding prototype symbols. To keep the compression visually lossless, we use a multi-stage symbol clustering procedure to group similar symbols and to ensure that there is no visible error in the decompressed image. In the encoding phase, the resulting data streams are encoded by adaptive arithmetic coding. Our results show that the average compression ratio is better than the international standard JBIG2 and the compressed form of a document image is suitable for a content-based keyword searching operation.